dynamical modeling and distributed control of connected...

IEEE IntEllIgEnt transportatIon systEms magazInE • 46 • fall 2017 1939-1390/17©2017IEEE

Shengbo Eben Li and Yang ZhengContributed equally to this work.

Correspondence should be sent to K. Li. This research was supported by NSF China

under grant 51575293 and 51622504, and National Key R&D Program in China

with grant 2016YFB0100906. The work of Y. Zheng is supported by

the Clarendon Scholarship and the Jason Hu Scholarship.

Shengbo Eben Li and Keqiang LiThe State Key Lab of Automotive Safety and Energy, Department of Automotive

Engineering, Tsinghua University, Beijing 100084, China. E-mail: [email protected],

[email protected]

Yang ZhengThe Department of Engineering Science,

University of Oxford, Parks Road, OX1 3PJ, U.K. E-mail: [email protected]

Yujia Wu and J. Karl HedrickThe Department of Mechanical Engineering at UC Berkeley, Berkeley, CA, USA, 94720.

E-mail: [email protected], [email protected]

Feng GaoSchool of Electrical Engineering,

Chongqing University, Chongqing, China, 400044. E-mail: [email protected]

Hongwei ZhangThe Department of Computer Science,

Wayne State University, Detroit, MI 48202. E-mail: [email protected]

Digital Object Identifier 10.1109/MITS.2017.2709781Date of publication: 25 July 2017

Dynamical Modeling and Distributed Control of Connected and Automated

Vehicles: Challenges and Opportunities

©is

toc

kp

ho

to.c

om

/ch

om

bo

sa

n

Abstract—The platooning of connected and automated vehicles (CAVs) is expected to have a transformative im-pact on road transportation, e.g., enhancing highway safety, improving traffic utility, and reducing fuel con-sumption. Requiring only local information, distributed control schemes are scalable approaches to the coordi-nation of multiple CAVs without using centralized com-munication and computation. From the perspective of multi-agent consensus control, this paper introduces

Special Section on the IEEE-IV’15 Symposium

IEEE IntEllIgEnt transportatIon systEms magazInE • 47 • fall 2017

I. Introduction

The platooning of connected and automated vehicles (CAVs) on the highway has attracted extensive inter-est due to its potential to significantly impact road transportation. The control of a platoon aims to en-

sure all the vehicles in the same lane move at a consistent speed while maintaining the desired spacing between ad-jacent vehicles. To our best knowledge, the earliest imple-mentations date back to the PATH program in the 1980s, where many topics were studied such as division of control tasks, the layout of control architecture, as well as control laws for headway control [1]. Recently, some demos were performed in the real world, including the GCDC in the Netherlands [2], SARTRE in Europe [3], and Energy-ITS in Japan [4].

Earlier studies on platooning often only consider radar-based sensing systems, in which the types of information exchange topologies are quite limited. However, the rapid deployment of vehicle-to-vehicle (V2V) communications, such as dedicated short range communication (DSRC)[5], [6], can generate a variety of new topologies for platoons, e.g., two-predecessor following type and multiple-prede-cessor following type [7], [8]. New challenges naturally arise due to the topological variety, which is even critical when considering time delay, packet loss, and quantization error in the communications. In such cases, it is prefer-able to view a platoon as a network of dynamical systems, and to employ multi-agent consensus control schemes to design distributed controllers [7], [23]. For example, Wang et al. introduced a weighted and constrained consensus seeking framework to study the influence of time-varying network structures on the platoon dynamics by using a discrete-time Markov chain [10]. Bernardo et al. analyzed vehicle platoons from the viewpoint of consensus control of a dynamic network [11]. Zheng et al. introduced two funda-mental methods to improve the stability margin of platoons via topological selection and control adjustment [22]. Gao et al. proposed an H3 control method to address uncertain vehicle dynamics and time-delays [42]. More recently, both robustness analysis and distributed H3 control synthesis

have been discussed for a platoon of connected vehicles with undirected topologies in [12].

From the perspective of multi-agent consensus control, a platoon of CAVs is actually a one-dimensional network of dynamical systems, where the vehicles only use their neighboring information for feedback. This perspective naturally decomposes a platoon of CAVs into four interre-lated components, i.e., node dynamics (ND), information flow network (IFN), distributed controller (DC), and for-mation geometry (FG), which is originally proposed in [22], [23]. This decomposition is able to provide a unified frame-work to analyze, design, and synthesize the platoon sys-tem, as well as further on-road implementations [23], [24]. The main contributions of this paper are: 1) we summarize the modeling techniques of each component in a platoon according to the proposed four-component framework; 2) we present a detailed discussion of techniques on four per-formance metrics in a platoon, i.e., internal stability, stabil-ity margin, string stability, and coherence behavior; 3) we introduce basic distributed synthesis methods, including linear consensus control, distributed robust control, dis-tributed sliding mode control, and distributed model pre-dictive control.

The rest of this paper is organized as follows: Section II introduces the four-component framework, and presents the modeling techniques of each component. Section III reviews four main performance metrics, followed by dis-cussions on distributed controllers in Section IV. Section V concludes this paper.

II. Modeling of a platoon of CAVs: the Four-component FrameworkThis paper considers a platoon of CAVs on a flat road, which aims to move at the same speed while maintaining desired spacing among vehicles. The platoon has a leading ve-hicle and other following vehicles. As shown in Fig. 1, the platoon system can be viewed as a combination of four main components [22], [23]:1) Node dynamics (ND), which describes the behavior of

each involved CAV;

a decomposition framework to model, analyze, and design the platoon system. In this framework, a platoon is naturally decomposed into four interrelated components, i.e., 1) node dynamics, 2) information flow network, 3) distributed controller, and 4) geom-etry formation. The classic model of each component is summarized according to the results of the literature survey; four main performance metrics, i.e., internal stability, stability margin, string stability, and coherence behavior, are discussed in the same fashion. Also, the basis of typical distributed control techniques is presented, including linear consensus control, distributed robust control, distributed sliding mode control, and distributed model predictive control.


2) Information flow network (IFN), which defines how the nodes exchange information with each other, including the topology and quality of information flow;

3) Distributed controller (DC), which implements the feedback control only using neighboring information;

4) Formation geometry (FG), which dictates the desired inter-vehicle distance when platooning.Each component in Fig. 1 has significant influence on

the collective behavior of a platoon. According to the four-component framework, a categorization of existing litera-ture can be found in [23], [24]. In this section, we present a detailed discussion of the modeling techniques for each component. For completeness, we first introduce the defi-nitions of some performance measures that are widely used in platoon control.

Definition 1 (Internal Stability). A linear platoon is said to be internally stable if and only if the closed-loop sys-tem has eigenvalues with strictly negative real parts [5], [7];

Definition 2 (Stability Margin). The stability margin of a platoon is defined as the absolute value of the real part of the least stable eigenvalue, which characterizes the con-vergence speed of initial errors [22], [25];

Definition 3 (String Stability). A platoon is said to be string stable if the disturbances are not amplified when propagated downstream along the vehicle string [27], [33];

Definition 4 (Coherence Behavior). The coherence be-havior describes how well the formation resembles a rigid body subject to exogenous disturbances [39], [54], which is quantified as a certain H2 norm of the closed-loop system.

Note that we only present the ba-sic descriptions of the performance metrics, and the exact mathematical definitions might be slightly differ-ent in the literature.

A. Node Dynamics (ND)Many previous studies on platoon control only emphasize on the longi-tudinal dynamical behaviors. Only a few studies discussed the integrated longitudinal and lateral control [44], [45]. Bicycle model is usually used to describe the lateral dynamics for con-trol design (see [44], [45] for details). Here, we only review the modeling of longitudinal dynamics. In addition, we mainly focus on continuous mod-els, and there is an alternative class of modeling techniques using timed automata or hybrid models; see [65]–[67] for details.

Vehicle longitudinal dynamics are inherently nonlinear, consisting of drive line, brake system, aerodynamics drag,

rolling resistance, gravitational force [15], [29]. The following nonlinear equations and its variants are widely employed to model the nonlinear longitudinal dynamics:

, ,

p t v t

r T t m v t C v t m gf

T t T t T t

i N,

,,

,

i i

w i

T ii i i A i i i

i i i i

2

des

!h

x

=

= + +

+ =

o

o

o

^

^

^

^

^

^

^

^

h

h

h

h

h

h

h

h

Z

[

\

]]

]] (1)

where, , , , ; N p t v t1 2 andN i if= ^ ^h h" , denote the posi-tion and velocity of vehicle ;i mi is the vehicle mass; C ,A i

is the lumped aerodynamic drag coefficient; g is the ac-celeration due to gravity; f is the coefficient of rolling re-sistance; T ti ^ h denotes the actual driving/braking torque; T t,ides ^ h is the desired driving/braking torque; ix is the in-ertial delay of vehicle longitudinal dynamics; r ,w i denotes the wheel radius and ,T ih is the mechanical efficiency of driveline. The position and velocity of the leading vehicle are denoted by p t0 ^ h and ,v t 0 ^ h respectively.

Some studies directly use nonlinear models for platoon control (see [19], [21], [27], [34], [45]). The asymptotic stability and string stability can be guaranteed by carefully selecting the control parameters, but explicit performance limits are rather difficult to analyze with given spacing policy and communication topology. Actually, linear models are more frequently used for tractable issues. In the literature, the commonly used models include 1) single integrator model, 2) second-order model, 3) third-order model, and 4) single-input-single-output (SISO) model.

Formation Geometry

i

i

N

NHeterogeneity

Upper Bound

Lower Bound

0

CC

Information FlowTopology

C

Node Dynamics

DistributedController

xN

∆N∆i ∆i

∆i–1

xi

dr

ddes

v0

x0

tuN ui ui–1

i–1

i–1

xi–1

FIG 1 Four major components of a platoon [22], [23]: 1) node dynamics, 2) information flow network, 3) distributed controller, 4) geometry formation; where dr is the actual relative distance, ddes is the desired distance, ui is the the control signal, xi is the state, ∆ i denote the dynamical uncertainty, and C denotes the controller.

IEEE IntEllIgEnt transportatIon systEms magazInE • 48 • fall 2017 IEEE IntEllIgEnt transportatIon systEms magazInE • 49 • fall 2017

The single integrator model is the simplest case, which takes the vehicle speed as control input and position as the exclusive state, i.e.,

.p t u ti i=o ^ ^h h (2)

where the control input u t i^ h is the velocity of each vehicle. This single integrator model (2) can significantly simplify the theoretical analysis on controller design. For instance, the structured optimal con-trol of platoons can be transformed into a convex problem under the single integrator assumption, but this problem is challenging for other models [39]. However, in addition to largely departing from actual vehicle dynamics, the single integrator model fails to reproduce string instability [35]. An improvement is to assume ND as a point mass, resulting in the double-integrator model [18], [35], [36]:

.p t v tv t u t

i i

i i

=

=

o

o ^^

^^

hh

hh) (3)

where u t i^ h is the acceleration of each vehicle. Many im-portant theoretical results, like decentralized optimal control [39], stability margin analysis [18], [25], [26], and coherence behavior [54], rely on the assumption of second-order dynamics. This assumption still does not catch many features of real vehicle dynamics, e.g., inertial delay in powertrain dynamics. One modeling trend is to further in-crease one state and yield so-called third-order model. The added state is often to approximate the input/output behav-iors of powertrain dynamics, which equivalently degrades the control input to engine torque and/or braking torque [15], [17], [20]. Most approximations use either feedback linearization technique [5], [15], [37] or lower-layer control technique [1], [45], resulting in a state space model as:

x t A x t B u ti i i i i= +o ^ ^ ^h h h

, , ,x tpva

A B000

100

011

001

i

i

i

i

i

i

i

ix x

= =

-

=^ h

R

T

SSSS

R

T

SSSS

>V

X

WWWW

V

X

WWWW

H (4)

where , ,x t p v ai i i iT=^ h 6 @ is the state, u ti ^ h is the desired

acceleration of each vehicle, and ix is the time constant of approximated powertrain dynamics. An alternative of the abovementioned models is the transfer function model, which is often used to analyze string stability in frequen-cy domain.

, ,p s s v s v s s a s1 1i i i i= =^ ^ ^ ^h h h h

,a s H s u si i i=^ ^ ^h h h (5)

where , ,p s v s a si i i^ ^ ^h h h are the Laplace transforms of node i’s position, velocity and acceleration, respectively; u si^ h is the control input, and H si^ h is a linear single-in-put-single-output (SISO) strictly proper transfer function. This model has two integrators and a lower order iner-tial delay, which leads to some fundamental limitations for certain platoons [33]. The pioneer work on this model started from Seiler, Pant, and Hedrick [33], and later wide-ly employed in many other studies; see e.g., [50] and [53].

B. Information Flow Network (IFN)The control of the collective behavior of multiple CAVs is based on vehicles’ mutual awareness of their states (e.g., inter-vehi-cle distance and vehicle speed), which is achieved by inter-vehicle sensing and communication. As shown in Fig. 1, the information provided by inter-vehicle sensing and communi-cation serves as an important input to each local controller, thus having a significant impact on the collective behavior. Here, we briefly discuss the topology and quality of informa-tion flow, as well as a graph-based modeling approach.1) Information Flow Topology (IFT). The IFT captures

the connectivity of information exchange between ve-hicles, and affects the platoon behavior such as string stability [33], stability margin [22], [25], and coherence behavior [39], [54]. Early-stage platoon control is mainly based on radar-sensing, and a vehicle can only obtain information about its nearest neighbors, i.e., imme-diately preceding and following vehicles. In this case, feasible IFTs include the predecessor following (PF) and bidirectional (BD) topologies, as shown in Fig. 2(a) and (b) respectively. Nowadays, with V2V communica-tion via technologies such as IEEE 802.11p-based DSRC and the emerging 5G solutions, various IFTs become feasible since a vehicle can communicate with vehicles beyond its immediate surroundings [5], [7]. Typical to-pologies include the predecessor-following leader (PFL) type, bidirectional leader (BDL) type, two predecessor-following (TPF) type, and two predecessor-following leader (TPFL) type, as shown in Fig. 2(c)–(f) respective-ly. Since IFT has a significant impact on the behavior of a platoon, it is important to adapt IFT (e.g., by control-ling the transmission power of communications) based on the need of platoon control [49].

Many important theoretical results rely on the assumption of second-order dynamics which still does not catch many features of real vehicle dynamics, e.g., inertial delay in powertrain dynamics.


2) Quality of information exchange. Given a specific IFT, the quality of inter-vehicle sensing and communica-tion also impacts the platoon behavior. For instance, the accuracy of radar sensing impacts the robustness and performance of platoon control. Meanwhile, wire-less communication quality significantly impacts the safety and performance of platoon control [47], [48]. Therefore, it is important to take the sensing and wire-less communication quality into account in platoon con-trol. For wireless communication whose quality can be controlled by using mechanisms such as scheduling, power control, and rate control, it is important to con-sider the joint control of wireless communication and vehicle platoon [63]. Existing IEEE 802.11p-based inter-vehicle wireless communication does not enable pre-dictable control of co-channel wireless interference and thus unable to ensure predictable control of communi-cation quality. With recent breakthroughs in wireless control networking, wireless communication quality can be controlled in a predictable manner [49], thus having opened the door to the co-design of vehicular wireless networking and platoon control [63].

3) Graph-based topological modeling. Directed graphs can be used to model allowable information flow between the vehicles in a platoon [7], [22]. More descriptions on graph theory can be found in [59] and the references therein.The information flow among followers is described by

a directed graph , , G V E AN N N N= " , with a set of nodes , , , ,N1 2VN f= " , a set of edges E V VN N N#3 and

the adjacency matrix .a RAN ijN N!= #6 @ Each edge ,j i^ h

represents a directed information flow from j to .i An edge ,j i^ h belongs to ,EN if and only if a 1ij = ; other-wise .a 0ij = It is assumed that there are no self-edges. The neighbor set of node i is denoted by . j a 1Ni ij= =" ,

The in-degree of i-th node is .deg ai ijj

N

1=

=/ Denote

, , , ,deg deg degdiagDN n1 2 f= " , and the Laplacian ma-trix l RL ij

N N!= #6 @ of GN is defined as .L D AN N= -

To model the information flow from the leader to the followers, we define an augmented graph as GN 1+ with a set of nodes , , , , N0 1 2VN 1 f=+ " , and a set of edges

. E V VN N N1 1 1#3+ + + A pinning matrix represents how each follower connects the leader, which is defined as

, , , ,diagP N1 2/ / /f= " , where 1 i/ = if edge , ;i0 EN 1! +" , otherwise .0i/ = The leader accessible set of node i is

, if, if .0 1

0Pii

i/4

/=

=

=)" ,

A spanning tree is a directed path connecting all the nodes in the graph [59]. The augmented graph GN 1+ should contain one spanning tree rooted at the leader for control-lability. It is easy to see that all the IFTs demonstrated in Fig. 2 contain at least a spanning tree. To illustrate some notations, considering PF and BD topologies, we have

, ,

01 1

1 1

10

0

L PPF PFj j j

=-

-

=

R

T

SSSSSS

R

T

SSSSSS

V

X

WWWWWW

V

X

WWWWWW

, .

11

12

111

10

0

L PBD BDj

j

j j=-

-

-

-=

R

T

SSSSS

R

T

SSSSS

V

X

WWWWW

V

X

WWWWW

The matrices ,L P encapsulate the topological connec-tions of the information flow in a platoon. In the proposed framework, L P+ plays an important role in the closed-loop dynamics (see (15)). The eigenvalues of L P+ have

...

...

... ...

...

(a)

(c)

...(b)

(d)

(e) (f)

FIG 2 Typical IFTs: (a) PF, (b) BD, (c) PFL, (d) BDL, (e) TPF, (f) TPFL.


a key impact on the stability margin for platoons with lin-ear node dynamics (see Lemma 5 in [7]). It is proved that all the eigenvalues of L P+ locate in the open right-half plane when GN 1+ contains one spanning tree rooted at the leader [7], [59]. Further, it is shown that for undirected to-pologies, the minimum eigenvalue of L P+ has a close re-lationship with the number of followers that are pinned to the leader (see Theorem 1 in [22]). The discussion in this paper mainly focuses on the impact of IFT, but the graph-based modeling approach can be extended to model the quality of information flow by introducing weights to each edge of the graph [59].

C. Distributed Controller (DC)The DC implements the feedback control using neighbors’ information, specified by ,I N Pi i i,= to enable the global coordination. An unstructured DC is one that corresponds to a complete graph which requires all-to-all communica-tions. Many existing studies consider structured control laws either in an explicit or implicit way; see [14], [15], [37] and [39].

The commonly used DC is linear for comprehensive results on theoretical analysis, and convenience in hard-ware implementations [7]. The general form of linear con-troller is:

[

],

u t k p t p t d

k v t v tk a t a t

, ,

,

,

ij

ij p i ii j ij i j

ij v i ii j ij

ij a i ij j ij

Iic c

c c

c c

=- - - - -

+ - - -

+ - - -

!

/^

^

^

^

^

^ ^

^

^

^h

h

h

h

h

h

h

h

h h

(6)

where k ,#ij # , ,p v a=^ h is the local controller gain, iic is the time delay corresponding to obtain its own state, and

ijc is the time delay corresponding to receive the state of node j via a communication channel. Many previous work only employed specific types of (6). The internal stability of a platoon with a linear controller largely depends on the structure of IFT. For example, the stabilizing region of lin-ear control gains was explicitly derived in [7] for a class of topologies, and string stability requirements for PF to-pology were established in [15]. The optimization methods, either numerical or analytical, were also used to optimize the localized gains [36], [39]. There are also some studies employing sliding mode control (SMC) to design a string-stable platoon [15]. For SMC, the internal stability and string stability of platoons are usually realized through a posterior controller tuning.

There are two main drawbacks in the design methods above: 1) they are unable to explicitly address string sta-bility, and 2) they are unable to handle the state or con-trol constraints. Recently, H3 controller synthesis has been proposed to include the string stability requirement in the design specification [20]. Model predictive control (MPC) has been introduced into platoon control to fore-

cast system dynamics, explicitly handling actuator/state constraints by optimizing given objectives [2], [19], [21]. In Section IV, we will introduce more details on how to design linear consensus controller, distributed robust con-troller, sliding mode controller, and model predictive con-troller for platoons.

D. Formation Geometry (FG)The objective of platoon control is to track the speed of the leader and to maintain a desired formation governed by an inter-vehicle spacing policy, i.e.,

, ,limlim

v t v tp t p t d i

00 N

,

ti

ti i i i

0

1 1

∞

∞

!- =

- - ="

"- -

^^

^^

hh

hh) (7)

where d ,i i1- is the desired space between i 1- and ,i which determines the formation geometry of a platoon.

There are three major policies of FG: 1) constant dis-tance (CD) policy, 2) constant time headway (CTH) policy, and 3) nonlinear distance (NLD) policy [13]. For the CD policy, the desired distance between two consecutive ve-hicles is independent of vehicle velocity, which can lead to a high traffic capacity. In this case, d ,i i1- is a given con-stant number,

, ,d d i N,i i1 0 !=- (8)

where d0 is a positive number. For CTH policy, the de-sired inter-vehicle range varies with the velocity, which is in accord with driver behaviors to some extent but limits achievable traffic capacity. One commonly used formula-tion is:

, ,d t v d i N,i i h i1 0 != +- (9)

where th is the time headway. For NLD policy, the desired inter-vehicle is a nonlinear function of velocity, i.e.,

, .d g v i N,i i i1 !=- ^ h (10)

Note that NLD policy has the potential to improve both the traffic flow stability and traffic capacity compared with CD and CTH policies [58].

III. Performance of a Platoon of CAVs: Stability and RobustnessIn this section, we discuss the performance of a platoon of CAVs, with a special focus on stability and robustness. Some practical benefits, such as reducing fuel consump-tion and improving traffic efficiency, are not covered here (we refer the interested reader to [30]–[32]). For example, a special design for fuel optimized platoon control was designed in [32], which uses a distributed


Pulse-and-Glide (PnG) controller to switch the engine operation point between two optimum positions to achieve lower average fuel.

In the field of platoon control, there are four important and commonly discussed performance metrics, namely, internal stability, stability margin, string stability, and coherence behavior. In general, the first two metrics fo-cus on stability of a platoon, while the other two metrics focus on the robustness of a platoon considering external disturbances. We note that a detailed categorization can be found in [24]. In addition to these performance mea-sures, some other metrics, such as fast convergence and safety, are also important when designing a practical pla-toon system.

A. Internal StabilityNo matter what kind of topology is employed, internal stabil-ity must be guaranteed in a platoon. Two main approaches have been proposed to ensure the internal stability: 1) global approach [17], [57], and 2) local approach [9], [15], [21].

The first approach is to straightforwardly take the overall platoon as a structured system and then design a controller in a centralized way, in which IFT becomes less important in the design process. For example, a linear matrix inequality (LMI) was obtained based on the global platoon dynamics to guarantee internal stability [17], [57]. One major drawback of this approach is that the com-putation efficiency quickly worsens with increasing the platoon size. Therefore, most studies decomposed a pla-toon into sub-systems and applied decentralized control methods, leading to the second approach. For instance, under PF topology, a platoon can be naturally viewed as unidirectional cascade systems, which only needs to study any two successive vehicles to guarantee stability; see [9], [15], [20] and [28]. Besides, the inclusion principle was used to decompose such kind of platoon into local subsystems, where an overlapping controller was designed [37]. This technique is not suitable for a platoon with BD topology since its spacing errors propagate from both forward and backward directions. Partial differential equation (PDE) techniques were applied to approximate the dynamics of platoons with BD topologies [18], [25], [26], which could avoid the analysis of high dimensional dynamics. In ad-ditional, Lyapunov method based energy function was used to prove both longitudinal car following stability and latitudinal lane keeping stability [44]. As an extension, La-salle’s invariance principle can also be utilized to prove asymptotic stability for time invariant platoon systems in the case where the derivative of a Lyapunov candidate is only negative semi-definite.

B. Stability MarginStability margin is used to characterize the convergence speed of the spacing errors in a platoon [22]. Most of cur-

rent research on stability margin focus on the CD policy, which has revealed that stability margin is a function of 1) platoon size ( N ), 2) ND, 3) IFN, 4) DC structure [18], [22], [25], [26].

By considering ND as a point mass, Barooah et. al proved that the stability margin approached zero as /O N1 2^ h un-der symmetric bidirectional control, and asymptotic be-havior of stability margin could be improved to /O N1^ h by introducing small amounts of “mistuning” [18]. This result was extended to linear third-order dynamics, which cov-ers the inertial delay of powertrain dynamics in [7]. Using the PDE approximation, Hao et al. showed that the scal-ing law of stability margin could be improved to /O N1 /D2^ h under D-dimensional IFTs [25]. Recently, it was shown that employing asymmetric control, the stability margin could be bounded away from zero, which is independent of the platoon size [26]. Zheng et al. further pointed out two basic methods to improve the stability margin via topology selec-tion and control adjustment [22].

C. String StabilityThe achievability of string stability has a tight relationship with the FG and IFN in a platoon. Seiler et al. showed that string stability cannot be guaranteed for any linear identi-cal controllers under PF topology and CD policy [33]. Ba-rooah et al. further pointed out that for a homogeneous platoon with BD topology, linear identical controllers also suffered fundamental limitations on the string stability [50]. Middleton et al. extended the work in [33] by consider-ing heterogeneous ND, limited communication range, non-zero time headway policy, which showed that both forward communication range and small time headway cannot alter the string instability [53]. Some solutions have been proposed to improve string stability, including: 1) relax-ing formation rigidity, i.e., introducing enough time head-way in the spacing policy [14], [15], [36], or using nonlinear policy [58]; 2) using non-identical controllers for different vehicles [18], [38]; 3) extending the information flow by us-ing more complex IFTs [16], [37].

Recently, some advanced controllers have been pro-posed to ensure string stability, including sliding mode con-trol [15], model predictive control [2], [19] and H3 control [17], [20]. Note that all of them either employ CTH policy or use the leader’s information. Current research results usu-ally focus on string stability caused by disturbances or ma-neuvers of the lead vehicle. However, following vehicles in a platoon also have the same probability encountering ex-ternal disturbances. There are a few research results that consider the effects of disturbances from any vehicle in a platoon. For example, Seiler et al. [33] investigated multiple disturbances propagate in the platoon, where string insta-bility was addressed based on a constant spacing. This work has been extended to a more practical situation with a variable spacing strategy in [52].


D. Coherence BehaviorThe coherence behavior is a scalar metric adopting H2-norm of the closed-loop system, which characterizes the robustness of a platoon driven by exogenous disturbances. This captures the notion of coherence [40], [54]. Bamieh et al. investigated the asymptotic scaling of upper bounds on coherence behavior with respect to the platoon size, and indicated that the IFN may play a more important role than DC [54]. Several recent research used coherence behavior as the cost function to optimize the local control gains using augmented Lagrangian approach [39]. In addi-tion, alternative direction method of multipliers (ADMM) was used to optimize the communication structure of IFTs in [40], where the cost function was the measure of coherence behavior. Recently, chordal decomposition has been applied in the design of structured controllers, which has the potential to address the coherence behav-ior of platoons efficiently [55]; also see one recent result in semidefinite programming [56].

IV. Controller Design of a platoon of CAVs: Distributed MethodsThis section introduces four types of distributed controller design methods for a platoon of CAVs, i.e., linear consensus control, robust control, distributed sliding mode control, and distributed model predictive control. For the first two methods, linear vehicle models are used, while nonlinear vehicle models can be employed for the last two methods. The main strategy of these methods is to decouple the pla-toon dynamics into several subsystems depending on the eigenvalues of .L P+ After decoupling, many synthesis methods can be used to solve the required distributed con-troller using LMIs.

A. Linear Consensus ControlLinear control is one of the most commonly used methods for platoon control [13], [16], [18], since it can not only facili-tate theoretical analysis but is also suitable for hardware implementations. Many existing results on stability region, stability margin, and string stability requirements [16], [17], [37] are based on linear controllers.

Here, we introduce a generic approach to analyze the collective behavior of platoons when employing linear consensus controllers (see [7], [22] for more details). The 3rd-order state space model is adopted to describe node dynamics, as shown in (4). The platoon is assumed to be homogeneous (i.e., , )A A i Ni != and the CD policy is em-ployed. The general linear consensus controller is given in (6). It is assumed that controller gains are identical, i.e., k k,# #i = # , , ,p v a=^ h .i N! Besides, we assume that there are no time-delays, i.e., , , .i j0 Nij !c = To write (6) into a compact form, we define a new tracking error

,x t x t x t di i i0= - -u u^ ^ ^h h h (11)

where , , .d d 0 0 ,i iT

0=u 6 @ For CD policy, the desired distance between the i-th follower and the leader is .d i d ,i 0 0#=- Then, (6) is rewritten into

u t k x t x t

ij

Ti j

Ii

=- -!

/ u u^ ^ ^ ^h h hh, (12)

where , ,k k k kp v aT= 6 @ is a vector of local feedback gains.

To derive the collective dynamics of a platoon, we define the collective state vector , , ,X x x x RT T

NT T N

1 23 1f != #u u u6 @ and the

collective control input vector , , , .U u u u RNT N

1 21f != #6 @

The collective dynamics of nodes from 1 to N are

X t I A X t I B U tN N$ $, ,= +o ^ ^ ^h h h (13)

with ,, I A I BR RNN N

NN N3 3 3! !, ,# # where , denotes

the Kronecker product. Based on (12), the collective form of the distributed control law is written into

.U t k X tL P T $,=- +^ ^ ^h h h (14)

Substituting (14) into (13), the closed-loop dynamics of a homogeneous platoon become

.X t I A Bk X tL PNT $, ,= - +o ^ ^ ^h h h6 @ (15)

Using the eigenvalue decomposition of ,L P+ it is proved that (15) is asymptotically stable if and only if

, .A Bk i NiT 6 !m- (16)

are all Hurwitz, where , i Ni !m are the eigenvalues of L P+ [7]. By this way, the collective dynamics of a pla-toon are reduced into the behavior of multiple subsystems. Thus, the design of feedback gains is decoupled from the IFT, leading to scalable solutions for large-scale platoons [7]. Note that the eigenvalue decomposition of L P+ plays a key role in decoupling the dynamics from IFTs, which is also used in the design of distributed robust control (see (19)).

As shown in (15), the closed-loop dynamics of a vehicu-lar platoon is a function of four components, namely 1) ND, denoted by , ;A B 2) IFT, denoted by L P+ ; 3) FG, included in X as the desired distance (see (11)); and 4) DC, denot-ed by .k In addition, it is easy to see that the performance measures of platoons must have a tight relationship with the four main components. However, it is usually rather difficult to explicitly obtain the relationship between the performance metrics and the decomposed components. Most of existing research on string stability focuses on spe-cific cases only; see [9], [16], [17] and [37] for example.

B. Distributed Robust ControlThe robustness of platoon control systems is an important topic. One practical way to handle model mismatches in


vehicle dynamics is to use the consistent and accurate in-put-output behavior of node dynamics [41]. But it is not easy to accommodate the heterogeneity in node dynamics.

Considering the requirements of string stability, robust-ness, and tracking performance, Gao et al. proposed an H3 control method for a heterogeneous platoon with uncertain dynamics and uniform time delays [42]. In this study, all nodes were combined as a big system. One disadvantage is that the designed controller only works for a specific platoon and it needs to be redesigned when the scale or interaction topology changes. Similar to the case of linear control, the decoupling strategy of robust control is also an effective way to overcome this problem. This strategy is motivated by the eigenvalue decomposition of ,L P+ and can balance the performances of robustness and distur-bance attenuation. As shown in Fig. 3, the coupling arising from information topology can be successfully transferred to the uncertain parts of nodes V by applying a linear trans-formation to the platoon.

In Fig. 3, the dynamics of a platoon in frequency domain are derived by adding the model uncertainties to (5):

,E ss

H sU s

sW s p s s

1 11N2 2 0 0C= + - -^^

^ ^ ^hh

h h h

,Z s s P s U sΩ=^ ^ ^ ^h h h h ,W s Z sV=^ ^h h (17)

where /

/, ,E s

e s

e s

p s p s d s

p s p s d s

1

11

,

,N N N

N

1 1 0 1 0

0 0

h h h= =

- -

- -

=^

^

^

^

^

^

^

h

h

h

h

h

h

h> > >H H H

, , ,d

ddiag

,

,N

N0

1 0

0

1h fT TVC = = ^ h> H is the normalized model

uncertainty satisfying ; ( )s1iT # X3 is its weighting function; W s^ h is the disturbance arising from ,V whose input is .Z s^ h Note that this model (17) includes additional uncertainty part iT compared to the 3rd-order state space model (4). The CD policy is used here. Each

node is controlled by the distributed state-feedback con-trol logic (12). Here, we present its expression in frequen-cy domain:

,U s K s E sL P= +^ ^ ^ ^h h h h (18)

where .K s k k s k sp v a2= + +^ h The nodes are interacting

by the information topology .L P+ For undirected topolo-gies, L P+ has an eigenvalue decomposition as

, , .diagL P N11fm mW W+ = -^ h (19)

where im is the eigenvalue of L P+ , and W is the eigenvector matrix of L P+ . Using the linear transformation in Fig. 3, the certain parts are then decoupled to a diago nal structure:

,E ss

H sU s

sW s p s s

1 11N2 21

0 0CW= + - --r r rr ^^

^ ^ ^hh

h h h

, , ,U s K s diag E sN1 fm m=r r^ ^ ^ ^h h h h

Ω , .Z s s P s U s W s Z sV= =r r r rr^ ^ ^ ^ ^ ^h h h h h h (20)

The variables in (20) are linear transformation of its original counterparts:

, ,Z s Z s U s U s1 1W W= =- -r r^ ^ ^ ^h h h h

, ,E s E s W s W s1 1W W= =- -r r^ ^ ^ ^h h h h

, .01

01V VCC W W W= =- -r r r (21)

Based on (20), the synthesis approach of H3 control can be used to numerically solve K s^ h offline, which ensures the requirements of robustness and disturbance.

C. Distributed Sliding Mode ControlThe sliding mode control (SMC) is a promising method for platooning of multiple vehicles to handle nonlinear dynamics

and actuator saturations. In [27], an adaptive SMC was proposed for equi-librium-stable interconnected systems to guarantee string stability. In this study, however, applicable topologies are limited to unidirectional topolo-gies, where one node can only obtain the information from its predeces-sors. In [64], an urban scenario pla-toon control method was introduced with two basic modes: cruise mode and collision avoidance mode.

In this section, we introduce a dis -tributed SMC scheme for platoons with homogeneous linear dynamics (4) and undirected (symmetric) IFTs.

W (s ) Ψ –1W (s )

U (s ) Ψ –1U (s )

Z (s )

E (s )

Ψ –1Z (s )

Ψ –1E (s )

diag(∆1, . . . , ∆N)

H (s ), Ω(s )

K (s ) (L+ P )

Ψ –1 ∆Ψ

H (s ), Ω(s )

K(s).diag(λ1, . . . , λN)

Coupled Decoupled

LinearTransformation

FIG 3 Decoupling of vehicular platoon systems.


If there exists a spanning tree in GN 1+ and information flow between followers is undirected, L P+ is positive definite [22]. The design of distributed SMC is divided into two stag-es, i .e., 1) topological sliding surface design and 2) to-pological reaching law design. The design of sliding surface and reaching law highly depends on the following topological structured function:

,Z a z z p z,

ij j i

N

ij i j i i1

C = - +!=

/^ ^h h

where , , , , , ,Z z z z i j NN1 2 f_ !6 @ aij and Pi are elements of the adjacency matrix and pinning matrix. Using topo-logical structured function, the distributed sliding sur-face is designed as

s t c X p c xii PT

i PT

oC= -^ ^h h

,a c x c x p c x c x,j j i

N

ij PT

i PT

j i PT

i PT

o1

= - + -!=

/ ^ ^h h

where , , , c X c x x xPT

PT

N1 2 f= 6 @ and c Rp3 1! # is the com-

mon vector coefficient of distributed sliding mode, satisfy-ing that c B P

Ti is invertible. Note that (4) is a single input

model, and the invertibility of c BPT

i means that c BPT

i is a nonzero constant. The topological sliding surface for the whole system is defined as

.Ss

s

c x x

c x xL P

N

PT

PT

N

1 1 0

0

h h_ =

-

-

+^

^

^h

h

h> >H H

The topological reaching law is designed by substitut-ing sim- to topological structured function

,s s a s s p s,

i i ij j i

N

ij i j i i1

_ m mC - =- - +!=

/o ^ e ^h h o

where 02m is the common coefficient for the distributed re -aching law. The topological reaching law for the whole system is

.S t SL Pm=- +o ^ ^h h

Comparing the derivative of sliding error and reaching law, we can cancel L P+ if it is positive definite,

, , .c x x c x x SL PL P pT

pT

NT

1 0 0f m- - =- ++ o o o o^ ^ ^ ^h h h h6 @ (22)

The positive definiteness of L P+ is essential since it guarantees the cancellation. Then, we have

.c A x c B u s pT

i i pT

i i im+ = - (23)

The control law is obtained by solving the upper equa-tion. The stability proof of distributed SMC is also divided

into two parts, i.e., reaching phase analysis and sliding phase analysis. The stability of the reaching phase is ana-lyzed with respect to the whole platoon by choosing a Ly-apunov candidate

.V t S S21 L PT 1= + -^ ^h h (24)

By taking the derivative of Lyapunov function, it can be shown that sliding surface is reached asymptotically. Dur-ing the sliding phase, a proper selection of cp is necessary to achieve stable sliding dynamics. Details can be found in [46].

D. Distributed Model Predictive ControlModel predictive control (MPC) is an optimization-based control technique to anticipate future behavior of plants and take control actions accordingly. Using MPC tech-niques, the control input is obtained by numerically opti-mizing a finite horizon optimal control problem where both nonlinearity and constraints can be explicitly handled. This technique has been embraced by many industrial ap-plications, for instance, collision avoidance and vehicle sta-bility [29], [60].

Currently, most MPCs are implemented in a centralized way, where all the control inputs are computed by assum-ing all the states are known. When considering a platoon system involving multiple vehicles, the centralized imple-mentation is not suitable due to the challenges of gathering the information of all vehicles and solving a large-scale optimization problem. Most existing work for platoon con-trol relies on the problem formulation of adaptive cruise control (ACC) [61], [62], which only involves two vehicles in the problem formulation. There exist some extensions to the cooperative ACC which considers multiple vehicles [2], [19]. The treatments in [2] and [19], however, also directly consider two consecutive vehicles in the problem formula-tion, which are only applicable to limited types of commu-nication topologies.

Here, a synthesis method of distributed MPC is pre-sented for a heterogeneous platoon, where each vehicle is assigned a local optimal control problem only relying on its neighboring vehicles’ information [21]. This method is suitable for any type of IFTs, and the asymptotical stability of the closed-loop platoon system can be derived for unidi-rectional topologies [21]. The discrete version of nonlinear equations (1) are used to model the longitudinal dynamics of each vehicle, i.e.,

,x t x t u t1i i i i i$z + = +^ ^ ^ ^h hh h

y t x ti ic=^ ^h h (25)

where , , ( / ) ,t0 0 1 Ri iT 3 13 ! x= #6 @ ,

10

01

00 R2 3!c = #; E

x Ri i3 1!z #^ h is defined as


( ) ( )

( ) ( ) ( )

( ) ( )

.

s t v t t

v t mt

R T t C v t m g f

T t T t t1,

,, ,i

i i

ii i

T ii A i i i i

ii

i

2

vehveh

T

T

T

zh

x

=

+

+ - -

-

c m

R

T

SSSSSS

V

X

WWWWWW

Now, we define the local open-loop optimal control prob-lem for each node i :

Problem :Fi For , , ,i N1 2 f! " , at time t

, , ,min J y u y y , ,u t u N t

i ip

ip

ia

ia

0 1ip

ip

pf --^

^ ^h

h h

,l k t k t k t k tk

N

i ip

ip

ia

ia

0

1p

==

-

-/ , , ,y u y y` ^ ^ ^ ^h h h h j (26)

subject to

,

, , ,

x k t x k t u k t

y k t x k t x t x t u k t

1

0 U

ip

i ip

i ip

ip

ip

ip

i ip

i

$

$ !

z

c

+ = +

= =

^

^

^

^

^

^

^

^ ^

h

h h

hh

h

h

h h

,y N t y N t d1I

,ip

pi j

ja

p i jIi

= +!

/ u^ ^ ^h h h

,T N t h v N tip

p i ip

p=^ ^ ^h hh (27)

where , ,u t u N t0 1ip

ip

pf -^ ^h h6 @ denotes the unknown variables, Np is the predictive time horizon, Ui denotes the convex set of input constraints, Ii is the cardinality of

,Ii and ,d d 0, ,i j i jT=u 6 @ denotes the desired distance vector

between i and ,j the function li in (26) is the cost associ-ated with node ,i y k ti

a ^ h is the assumed trajectory of node i, and y k ti

a- ^ h is the assumed trajectory of the neighbors.

The formulation of problem Fi only needs the informa-tion from its neighbors, and thus it is suitable for various communication topologies, including all of those shown in Fig. 2. The key idea in the design of each local optimal control problem Fi is to construct and transmit assumed trajectories: each vehicle solves a local optimal control problem to obtain its own control input, and then sends its assumed output trajectory to its neighbors. Here, the as-sumed variable is a shifted optimal result of the last-step problem ,Fi synthesized by disposing the first value and adding a last value. The last added value ensures that the vehicle moves at a constant speed. Details of this design and the asymptotical stability of the closed-loop system can be found in [21].

V. ConclusionThis paper has presented a four-component framework to model, analyze, and synthesize a platoon of CAVs from the perspective of multi-agent consensus control. This framework is well-suited for designing distributed con-trol schemes of CAV platooning. With the four-component framework, this paper has introduced the modeling tech-niques, discussed the major performance metrics, and pre-sented the design of four types of distributed controllers.

There are some open questions, especially considering the emerging V2V and V2I communications. Two of them are briefly discussed here: 1) How to analyze and synthe-size a platoon in a systematic fashion, considering the non-linearity of node dynamics, the variety of topologies, and the need for low-cost controllers? Most existing work only considers simplified models and specific topologies for controller synthesis. Communication issues, such as time delays, quantization errors, and packet loss pose a signifi-cant challenge to platoon control as well. 2) How to balance different performance metrics in platoon control when considering practical requirements from highway opera-tions? The balance of stability (e.g., internal stability and stability margin) and robustness (e.g., string stability and coherence behavior) for a platoon is attracting research interest. The ultimate objectives of platooning are to en-hance highway safety, improve traffic utility, and reduce fuel consumption. How to explicitly take practical perfor-mance requirements into account is rather challenging for platoon control.

AcknowledgmentThis article is in memory of Prof. J Karl Hedrick who has dedicated valuable time to contributing to this piece of work while he was fighting with cancer. His intellectual contribution to nonlinear control and transportation en-gineering and his dedication to students and colleagues have always inspired and will continue inspiring us to explore the unknown and educate the next-generation of scholars.

About the AuthorsShengbo Eben Li received the M.S. and Ph.D. degrees from Tsinghua University in 2006 and 2009. He is currently the associate professor at Tsinghua University. His active re-search interests include autonomous vehicle control, learning-based driver

assistance, energy management for electrified powertrain, distributed control and optimal estimation, etc. He is the author of over 100 journal/conference papers, and the co-inventor of over 20 patents. Dr. Li was the recipient of Ex-cellent Young Scholar of NSF China (2016), Yangtze River Scholar –Excellent Young Professor (2016), etc.

Yang Zheng received his B.S. and M.S. degrees from Tsinghua Univer-sity, Beijing, China, in 2013 and 2015, respectively. He is currently working toward the DPhil. degree in the De-partment of Engineering Science at the University of Oxford. His research

interests include distributed control of dynamical systems,


with applications on vehicular platoons. Mr. Zheng re-ceived the Best Student Paper Award at the 17th Interna-tional IEEE Conference on Intelligent Transportation Systems, in 2014, and Best Paper Award at the 14th Intelligent Transportation Systems Asia-Pacific Forum in 2015. He was the recipient of the National Scholarship and Outstand-ing Graduate in Tsinghua University, the Clarendon Schol-arship at the University of Oxford.

Keqiang Li obtained the M.S. and Ph.D. degrees from Chongqing University of China in 1988 and 1995 respectively and his B.Tech degree from Tsinghua University of China in 1985. Dr. Li is the Professor of Automotive Engineer-ing at Tsinghua University. His main

research areas include automotive control system, driver as-sistance system, and network/networked dynamics and con-trol. Currently he is leading the national key project on connected and automated vehicles in China. Dr. Li has been a recipient of “Changjiang Scholar Program Professor”, National Award for Technological Invention in China, etc.

Yujia Wu received the B.S. degree in Automation Science and Electrical Engineering from Beihang Universi-ty, Beijing, China, in 2013. He is now perusing the Ph.D. degree in Mechan-ical Engineering from University of California at Berkeley, USA. His re-

search interests are in the area of nonlinear dynamical systems analysis and control design; distributed motion coordinate for groups of autonomous agents.

J. Karl Hedrick received the B.S. de-gree in engineering mechanics from University of Michigan, Ann Arbor, MI, USA, in 1966 and the M.S. and Ph.D. degrees in aeronautical and as-tronautical engineering from Stan-ford University, USA, in 1970 and 1971,

respectively. He was the James Marshall Wells Professor of University of California, Berkeley, CA, USA. His research interests include nonlinear control and its application to transportation systems.

Feng Gao received M.S. and Ph.D. in Tsinghua University in 2003 and 2007, respectively. He is now a profes-sor in School of Electrical Engineering, Chongqing University. His current re-search interests include robust con-trol and optimization approach with

application to automotive systems. He is the author of more

than 40 peer-reviewed journal and conference papers, and co-inventor of over 20 patents in China.

Hongwei Zhang (S’01-M’07-SM’13) received his B.S. and M.S. degrees in Computer Engineering from Chongq-ing University, China and his Ph.D. degree in Computer Science and Engi-neering from The Ohio State Uni-versity. He is currently an associate

pro fessor of computer science at Wayne State University. His primary research interests lie in cyber-physical systems, wireless sensing and control networks, and dependable net-worked systems. He is a recipient of the NSF CAREER Award. His work has won the Best Demo Award at the 21st NSF GENI Engineering Conference in 2014.

References[1] S. Shladover, C. Desoer, J. Hedrick, M. Tomizuka, J. Walrand, W.

Zhang, D. McMahon, H. Peng, S. Sheikholeslam, and N. McKeown, “Automated vehicle control developments in the PATH program,” IEEE Trans. Vehicular Tech, vol. 40, no. 1, pp. 114–130, 1991.

[2] R. Kianfar, B. Augusto, A. Ebadighajari, et al. “Design and experimen-tal validation of a cooperative driving system in the grand cooperative driving challenge,” IEEE Trans. Intell. Transp. Syst., vol. 13, no. 3, pp. 994–1007, 2012.

[3] T. Robinson, E. Chan, E. Coelingh “Operating platoons on public mo-torways: An introduction to the sartre platooning programme,” in Proc. 17th World Congress on Intelligent. Transportation System, Oct. 2010.

[4] S. Tsugawa, S. Kato, K. Aoki, “An Automated Truck Platoon for Energy Saving,” in Proc. IEEE/RSJ Int. Conf. Intelligent Robots and Systems, San Francisco, Sept. 2011, pp. 4109–4114.

[5] Y. Zheng, S. Eben Li, J. Wang, etc., “Influence of information flow topology on closed-loop stability of vehicle platoon with rigid forma-tion,” in Proc. 17th Int. IEEE Conf. ITSC, 2014, pp. 2094–2100.

[6] T. Willke, P. Tientrakool, and N. Maxemchuk, “A survey of inter-vehi-cle communication protocols and their applications,” IEEE Commun. Surv. Tut., vol. 11, no. 2, pp. 3–20, 2009.

[7] Y. Zheng, S. Li, etc. “Stability and scalability of homogeneous vehicu-lar platoon: Study on influence of information flow topologies,” IEEE Trans. Intell. Transp. Syst., vol. 17, no. 1, pp. 14–26, 2016.

[8] S. E. Li, Y. Zheng, K. Li, J. Wang, “Scalability limitation of homoge-neous vehicular platoon under undirected information flow topology and constant spacing policy,” in Proc. 34th Chinese Control Conf., Hang Zhou, July 28-30, 2015.

[9] S. Oncu, J. Ploeg, N. Wouw, and H. Nijmeijer, “Cooperative adaptive cruise control: Network-aware analysis of string stability,” IEEE Trans. Intell. Transp. Syst., vol. 15, no. 4, pp. 1527–1537, 2014.

[10] L. Wang, A. Syed, G. Yin, A. Pandya, and H. Zhang, “Control of vehicle platoons for highway safety and efficient utility: Consensus with com-munications and vehicle dynamics,” J. Syst. Sci. Complex., vol. 27, no. 4, pp. 605–631, 2014.

[11] M. Bernardo, S. Alessandro, and S. Stefania, “Distributed consensus strategy for platooning of vehicles in the presence of time-varying het-erogeneous communication delay,” IEEE Trans. Intell. Transp. Syst., vol. 16, no. 1, pp. 102–112, 2015.

[12] Y. Zheng, S. E. Li, K. Li, and W. Ren, “Platooning of Connected Vehicles with Undirected Topologies: Robustness Analysis and Distributed H-infinity Controller Synthesis,” arXiv:1611.01412. 2016.

[13] D. Swaroop, J. K. Hedrick, C. Chien, and P. Ioannou, “A comparison of spacing and headway control laws for automatically controlled ve-hicles,” Veh. Syst. Dyn., vol. 23, no. 1, pp. 597–625, 1994.

[14] G. Naus, R. Vugts, J. Ploeg, M. Molengraft, and M. Steinbuch, “String-stable CACC design and experimental validation: A frequency-domain approach,” IEEE Trans. Veh. Technol., vol. 59, no. 9, pp. 4268–4279, 2010.

[15] L. Xiao and F. Cao, “Practical string stability of platoon of adaptive cruise control vehicles,” IEEE Trans. Intell. Transp. Syst., vol. 12, no. 4, pp. 1184–1194, 2011.

[16] E. Shaw, J. K. Hedrick, “String stability analysis for heterogeneous ve-hicle strings,” Proc. Amer. Control Conf., 2007, pp. 3118– 3125.


[17] G. Guo and W. Yue, “Hierarchical platoon control with heterogeneous information feedback,” IET Control Theory Appl., pp. 1766–1781, 2011.

[18] P. Barooah, P. G. Mehta, and J. P. Hespanha, “Mistuning-based control design to improve closed-loop stability margin of vehicular platoons,” IEEE Trans. Autom. Control, vol. 54, no. 9, pp. 2100–2113, 2009.

[19] W. B. Dunbar and D. S. Caveney, “Distributed receding horizon control of vehicle platoons: Stability and string stability,” IEEE Trans. Auto. Control, vol. 57, no. 3, pp. 620–633, 2012.

[20] J. Ploeg, D. P. Shukla, and N. van de Wouw, “Controller synthesis for string stability of vehicle platoons,” IEEE Trans. Intell. Transp. Syst., vol. 15, no. 2, pp. 854–865, 2014.

[21] Y. Zheng, S. Eben Li, K. Li, et al. “Distributed model predictive control for heterogeneous vehicle platoons under unidirectional topologies,” IEEE Trans. Contr. Syst. Technol., vol. pp, no. 99, pp. 1–12, 2016.,

[22] Y. Zheng, S. Eben Li, K. Li, and L. Wang, “Stability margin improve-ment of vehicular platoon considering undirected topology and asym-metric control,” IEEE Trans. Contr. Syst. Technol., vol. 24, no. 4, pp. 1253–1265, 2016.

[23] Y. Zheng, “Dynamic modeling and distributed control of vehicular platoon under the four-component framework,” Master’s thesis, Tsin-ghua Univ., 2015.

[24] S. Eben Li, Y. Zheng, K. Li, J. Wang, “An overview of vehicular platoon control under the four-component framework,” In Proc. IEEE Intel-ligent Vehicles Symp., 2015, pp. 286–291.

[25] H. Hao, P. Barooah, and P. Mehta, “Stability margin scaling laws for distributed formation control as a function of network structure,” IEEE Trans. Autom. Control, vol. 56, no. 4, pp. 923–929, 2011.

[26] H. Hao and P. Barooah, “On achieving size-independent stability mar-gin of vehicular lattice formations with distributed control,” IEEE Trans. Autom. Control, vol. 57, no. 10, pp. 2688–2694, 2012.

[27] D. Swaroop and J. K. Hendrick, “String stability of interconnected sys-tems,” IEEE Trans. Autom. Control, vol. 41, no. 3, pp. 349–357, 1996.

[28] J. Ploeg, N. van Wouw, and H. Nijmeijer, “Lp string stability of cas-caded systems: Application to vehicle platooning,” IEEE Trans. Contr. Syst. Technol., vol. 22, no. 2, pp. 786–793, 2014.

[29] J. Wang, S. Eben Li, etc. “Longitudinal collision mitigation via coor-dinated braking of multiple vehicles using model predictive control,” Integr. Comput. Aided Eng., vol. 22, no. 2, pp. 171–185, 2015.

[30] A. Alam, A. Gattami, K. Johansson, “An experimental study on the fuel reduction potential of heavy duty vehicle platooning,” in Proc. 13th Int. IEEE Conf. ITSC, 2010, pp. 306– 311.

[31] B. van Arem, C. van Driel, and R. Visser, “The impact of cooperative adaptive cruise control on traffic-flow characteristics,” IEEE Trans. Intell. Transp. Syst., vol. 7, no. 4, pp. 429–436, 2006.

[32] S. Eben Li, K. Deng, Y. Zheng, and H. Peng, “Effect of pulse-and-glide strategy on traffic flow for a platoon of mixed automated and manually driven vehicles,” Comput. Aided Civil Infrastruct. Eng., vol. 30, no. 11, pp. 892–905, 2015.

[33] P. Seiler, A. Pant, and J. K. Hedrick, “Disturbance propagation in vehi-cle strings,” IEEE Trans. Autom. Control, vol. 49, no. 10, pp. 1835–1842, 2004.

[34] J. Kwon and D. Chwa, “Adaptive bidirectional platoon control using a coupled sliding mode control method,” IEEE Trans. Intell. Transp. Syst., vol. 15, no. 5, pp. 2040–2048, 2014.

[35] S. Darbha and P. Pagilla, “Limitations of employing undirected information flow graphs for the maintenance of rigid formations for heterogeneous vehicles,” Int. J. Eng. Sci., vol. 48, no. 11, pp. 1164–1178, 2010.

[36] C. Liang and H. Peng, “Optimal adaptive cruise control with guaranteed string stability,” Veh. Syst. Dyn., vol. 32, no. 4-5, pp. 313–330, 1999.

[37] S. Stankovic, M. Stanojevic, and D. Siljak, “Decentralized overlapping control of a platoon of vehicles,” IEEE Trans. Contr. Syst. Tech., vol. 8, no. 5, pp. 816–832, 2000.

[38] M. Khatir, E. Davidson, “Bounded stability and eventual string stabil-ity of a large platoon of vehicles using non-identical controllers,” in Proc. IEEE Conf. Decision Control, 2004, pp. 1111–1116.

[39] F. Lin, M. Fardad, and M. R. Jovanovic, “Optimal control of vehicular formations with nearest neighbor interactions,” IEEE Trans. Autom. Control, vol. 57, no. 9, pp. 2203–2218, 2012.

[40] F. Lin, M. Fardad, and M. R. Jovanović, “Design of optimal sparse feed-back gains via the alternating direction method of multipliers,” IEEE Trans. Autom. Control, vol. 58, no. 9, pp. 2426–2431, 2013.

[41] S. Eben Li, F. Gao, D. Cao, and K. Li, “Multiple model switching control of vehicle longitudinal dynamics for platoon level automation,” IEEE Trans. Veh. Technol., vol. 65, no. 6, pp. 4480–4492, 2016.

[42] F. Gao, S. Eben Li, et al. “Robust control of heterogeneous vehicular platoon with uncertain dynamics and communication delay,” IET Intell. Transport Syst., vol. 10, no. 7, pp. 503–513, 2016.

[43] J. A. Fax and R. Murray, “Information flow and cooperative control of vehicle formations,” IEEE Trans. Automat. Control, vol. 49, no. 9, pp. 1465–1476, 2004.

[44] Y. G. Liu and B. G. Xu, “Improved protocols and stability analysis for multi-vehicle cooperative autonomous systems,” IEEE Trans. Intell. Transp. Syst., vol. 16, no. 5, pp. 2700–2710, 2015.

[45] R. Rajamani, H. Tan, B. Law, and W. Zhang, “Demonstration of integrated longitudinal and lateral control for the operation of auto-mated vehicles in platoons,” IEEE Trans. Contr. Syst. Technol., 2000, pp. 695– 708.

[46] Y. Wu, S. E. Li, Y. Zheng, and J. K Hedrick, “Distributed sliding mode control for multi-vehicle systems with positive definite topologies,” in Proc. IEEE 55th Conf. Decision and Control, pp. 5213–5219. 2016.

[47] L. Xu, L. Wang, G. Yin, and H. Zhang, “Communication informa-tion structures and contents for enhanced safety of highway vehicle platoons,” IEEE Trans. Veh. Technol., vol. 63, no. 9, pp. 4206–4220, 2014.

[48] L. Xu, L. Wang, G. Yin, and H. Zhang, “Impact of communication era-sure channels on safety of highway vehicle platoons,” IEEE Trans. In-tell. Transp. Syst., vol. 16, no. 3, pp. 1456–1468, 2015.

[49] H. Zhang, X. Che, X. Liu, and X. Ju, “Adaptive instantiation of the pro-tocol interference model in wireless networked sensing and control,” ACM Trans. Sens. Netw., vol. 10, no. 2, pp. 28.1–28.48, 2014.

[50] P. Barooah and J.P. Hespanha, “Error amplification and disturbance propagation in vehicle strings with decentralized linear control,” in Proc. IEEE Conf. Decision Control, 2005, pp. 4964–4969.

[51] S. K. Yadlapalli, S. Darbha, and K. R. Rajagopal, “Information flow and its relation to stability of the motion of vehicles in a rigid formation,” IEEE Trans. Autom. Control, vol. 51, no. 8, pp. 1315–1319, 2006.

[52] Y. Liu, H. Gao, B. Xu, G. Liu, and H. Cheng, “Autonomous coordinated control of a platoon of vehicles with multiple disturbances,” IET Con-trol Theory Appl., vol. 8, no. 18, pp. 2325–2335, 2014.

[53] R. Middleton and J. Braslavsky, “String instability in classes of linear time invariant formation control with limited communication range,” IEEE Trans. Autom. Control, vol. 55, no. 7, pp. 1519–1530, 2010.

[54] B. Bamieh, M. Jovanovic, P. Mitra, and S. Patterson, “Coherence in large-scale networks: Dimension-dependent limitations of local feedback,” IEEE Trans. Autom. Control, vol. 57, no. 9, pp. 2235–2249, 2012.

[55] Y. Zheng, P. Mason, and A. Papachristodoulou, “A chordal decompo-sition approach to scalable design of structured feedback gains over directed graphs,” in Proc. IEEE 55th Conf. Decision and Control, 2016, pp. 6909–6914.

[56] Y. Zheng, G. Fantuzzi, A. Papachristodoulou, P. Goulart, and P. A. Wynn, “Fast ADMM for semidefinite programs with chordal sparsity,” arXiv preprint arXiv:1609.06068, 2016.

[57] G. Guo and W. Yue, “Autonomous platoon control allowing range- limited sensors,” IEEE Trans. Veh. Technol., vol. 61, pp. 2901–2912, 2012.

[58] K. Santhanakrishnan and R. Rajamani, “On spacing policies for high-way vehicle automation,” IEEE Trans. Intell. Transp. Syst., pp. 198–204, 2003.

[59] C. Godsil and G. Royle, Algebraic graph theory. New York: Springer-Verlag, 2001.

[60] S. Di Cairano, H. Tseng, D. Bernardini, and A. Bemporad, “Vehicle yaw stability control by coordinated active front steering and differential braking in the tire sideslip angles domain,” IEEE Trans. Control Syst. Technol., vol. 21, no. 4, pp. 1236–1248, 2013.

[61] P. Shakouri and A. Ordys, “Nonlinear model predictive control ap-proach in design of adaptive cruise control with automated switch-ing to cruise control,” Control Eng. Pract., vol. 26, pp. 160–177, 2014.

[62] S. Li, K. Li, R. Rajamani, and J. Wang, “Model predictive multi-ob-jective vehicular adaptive cruise control,” IEEE Trans. Control Syst. Technol., vol. 19, no. 3, pp. 556–566, 2011.

[63] G. Guo and S. Wen, “Communication scheduling and control of a pla-toon of vehicles in VANETs,” IEEE Trans. Intell. Transp. Syst., vol. 17, no. 6, pp. 1551–1563, 2016.

[64] A. Ferrara and C. Vecchio, “Second order sliding mode control of vehi-cles with distributed collision avoidance capabilities,” Mechatronics, vol. 19, no. 4, pp. 471–477, 2009.

[65] P. Varaiya, “Smart cars on smart roads: problems of control,” IEEE Trans. Autom. Control, vol. 38, no. 2, pp. 195–207, 1993.

[66] J. K. Hedrick, M. Tomizuka, and P. Varaiya, “Control issues in auto-mated highway systems,” IEEE Control Syst., vol. 14, no. 6, pp. 21–32, 1994.

[67] J. Lygeros, D. N. Godbole, and S. Sastry, “Verified hybrid controllers for automated vehicles,” IEEE Trans. Autom. Control, vol. 43, no. 4, pp. 522–539, 1998.

dynamical modeling and distributed control of connected...

Documents